Data source

We conducted a retrospective cohort study utilizing data from the South Korea National Health Insurance Service–National Health Insurance Database (NHIS–NHID). By utilizing unique health insurance card numbers shared within families and cross-referencing them with delivery dates, the NHIS accurately identified maternal-child pairs through a validated linkage algorithm [25,26]. The NHIS–NHID database provides near-complete coverage of the Korean population, as South Korea operates a single-payer, universal healthcare system in which nearly all residents (97%) are enrolled. This ensures that the data captures virtually all healthcare utilization in Korea, including inpatient, outpatient, and public health center services, minimizing selection bias related to healthcare access. Socioeconomic background is accounted for using income-level indicators derived from health insurance contributions, which serve as reliable proxies for income level. Additionally, given that Korea’s single-payer system ensures relatively uniform access to healthcare services, and supplementary policies targeting pregnant women and children further minimize disparities in provider access across income levels. The NHIS–NHID includes comprehensive data on sociodemographic factors, inpatient and outpatient healthcare services (including diagnoses, procedures, and medication prescriptions), health examination records for both mothers and their children, and vital statistics, which are linked with data from Statistics Korea. The requirement for informed consent was waived because personal codes are deidentified in the database. The protocol for this study was reviewed and approved by the institutional review board of Sungkyunkwan University (2024-04-060). All analyses were conducted in accordance with this protocol (S1 Protocol) and reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (S1 STROBE Checklist).

Cohorts for pregnancy and infancy analyses

Two distinct cohorts were constructed for all children born between April 1, 2009, and December 31, 2020. The first cohort to investigate antibiotic exposure during pregnancy (pregnancy analysis) included all children except for (1) those whose mothers were younger than 15 or older than 50 years, (2) those with chromosomal anomalies, (3) those whose mothers had taken antibiotics one month before but not during pregnancy to mitigate exposure misclassification, and (4) those whose mothers had no infection diagnosis during pregnancy. The second cohort to analyze antibiotic exposure in infancy and early childhood (infancy analysis) consisted of all children excluding (1) those who died within the first six months of life, (2) those with chromosomal anomalies, (3) those diagnosed with study outcomes within the first six months, and 4) those not diagnosed with an infection in the first six months of life (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Study flow chart for two distinctive cohort construction.

Abbreviations: NHIS, National Health Insurance Service.

https://doi.org/10.1371/journal.pmed.1004677.g001

Antibiotics exposure

Exposure was defined as one or more prescriptions of systemic antibiotic, using Anatomical Therapeutic Chemical classification system code, J01 (S2 Table). Since systemic antibiotics in South Korea can only be obtained with a physician’s prescription, all instances of antibiotic use were included and exposure misclassification from the over-the-counter was unlikely. In the pregnancy analysis, the antibiotic exposed group consisted of children whose mothers at least one antibiotic prescription during pregnancy. The non-exposed group included children whose mothers were not prescribed antibiotics from 30 days before last menstrual period (LMP) until the end of pregnancy. For the infancy analysis, the exposed group comprised children who received antibiotics at least once within the first six months of life. The non-exposed group consisted of children who did not receive any antibiotic prescriptions during the same timeframe.

Outcomes

The outcomes of interest included 6 autoimmune diseases including type 1 diabetes, juvenile idiopathic arthritis, ulcerative colitis, Crohn’s disease, systemic lupus erythematosus, and Hashimoto’s thyroiditis. All autoimmune disease events were required to have at least one inpatient or at least two outpatient diagnoses within one year for the improving specificity of the outcome (S2 Table). For the pregnancy analysis, children were followed from the day they were born until the occurrence of the outcome, death, or December 31, 2021. In the infancy analysis, children were followed from 180 days after birth until the occurrence of the outcome, death, or December 31, 2021.

Sociodemographic, clinical, and obstetric characteristics

For the analyses of both pregnancy and infancy, we assessed a broad range of potential confounders: (1) maternal demographic characteristics (e.g., maternal age, insurance type, and income level), (2) clinical characteristics such as cause of infections (e.g., respiratory infection, urinary infection), proxies for severity of infection (e.g., infection-related number of outpatient visits, distinct diagnoses), maternal comorbid conditions (e.g., asthma, migraine), maternal medication conditions (e.g., systemic corticosteroids, chronic obstructive pulmonary disease drugs), measures of healthcare utilization (e.g., number of out- or in-patient visits), and (3) obstetric characteristics including nulliparity, multiple gestations, obstetric comorbidity index [27]. For infancy analysis, we additionally considered sex of child, maternal exposure to antibiotics, preterm birth, cesarean section, birth weight, and type of feeding. Maternal sociodemographic characteristics were assessed at the time of delivery. Clinical characteristics were examined from six months prior to pregnancy up until the day before delivery, while causes of infections and proxies for severity of infection were measured during pregnancy or during early infancy, respectively. Lastly, measures of healthcare utilization were assessed during the 6 months prior to pregnancy but not during pregnancy. Maternal lifestyle-related factors, including smoking habits and body mass index (BMI) status were also described based on data from the national health screening program conducted before the LMP.

IPTW and sibling-matched analyses

We applied stabilized IPTW as our primary analytic approach to account for potential confounding. Although we initially planned to perform 1:1 propensity score matching using a greedy nearest neighbor algorithm without replacement, we revised our analytic strategy based on reviewer feedback and adopted IPTW to enhance statistical efficiency and generalizability, particularly given the comparable sizes of the antibiotics-exposed and unexposed groups. IPTW enabled the inclusion of a larger proportion of eligible subjects by assigning weights based on the probability of treatment, thereby reducing data loss while achieving covariate balance. We estimated PS, defined as the likelihood of antibiotic exposure in this study, through a logistic regression model that utilized all pre-defined covariates as independent variables. To prevent potential bias from extreme weights and to enhance exchangeability between two groups, we first performed 1% trimming within the overlapping region [28], followed by the application of stabilized IPTW. We assessed the balance of covariates between the exposed and unexposed groups using the standardized mean difference (SMD), considering values under 0.1 as indicative of balance [29,30]. We measured the incidence of outcomes per 100,000 person-years and calculated absolute rate differences with corresponding 95% confidence intervals (CIs) using Poisson regression. Hazard ratios (HRs) with 95% CIs were estimated using Cox proportional hazard regression models with robust standard errors to account for multiple pregnancies per mother.

As part of the main analysis, we also performed sibling-matched analyses restricted to children with siblings, including only sibling pairs who were discordant in exposure status, to account for potential unmeasured confounding by familial and maternal genetic factors. This approach inherently adjusts for time-invariant family-related characteristics. IRs and HRs for each outcome were estimated. HRs were derived using a stratified Cox proportional hazard regression model, with stratification by mother to account for familial clustering. The unit for all analyses was per child. Data management and analysis were conducted using SAS Enterprise Guide, version 7.1 (SAS Institute).

Subgroup and sensitivity analyses

Several subgroup analyses were conducted. First, we conducted a subgroup analysis according to broad- and narrow-spectrum antibiotics to evaluate whether the risk differed by antibiotic spectrum. Second, we analyzed the risk of study outcomes by subclasses among antibiotics which are frequently prescribed within database during pregnancy (cephalosporins, imidazoles, macrolides, penicillins) or infancy (cephalosporins, macrolides, penicillins), given that different effects were reported according to each antibiotic class. Third, we evaluated the cumulative duration-response relationship between antibiotic exposure and autoimmune disease. In this analysis, we identified the distribution of cumulative duration within each cohort and categorized them based on quartiles (<4, 4–7, 8–12, 13+ days for the pregnancy analysis and <8, 8–16, 17–28, +29 days for the infancy analysis). Fourth, considering the differential incidence of certain autoimmune diseases based on sex, we conducted subgroup analyses by the sex of the infant. Lastly, we examined whether the risk of autoimmune disease varied according to specific timing of exposure, delineated as the 1st, 2nd, and 3rd trimester for the pregnancy analysis, and 0 to <2, 2 to <4, and 4 to <6 months of age after birth for the infancy analysis.

Additionally, sensitivity analyses were conducted to examine the robustness of the main analysis. First, we modified the definition of exposure to two or more prescriptions of systemic antibiotics within the same assessment window, to minimize exposure misclassification. Second, given the stronger genetic predisposition for autoimmune diseases among twins, their inclusion may lead to overestimation of associations due to unmeasured confounding. To address this, we conducted a sensitivity analysis restricted to singleton births [31]. Third, we limited the study population to children who were breastfed to address the residual confounding from not breastfeeding, given that breastfeeding plays a significant role in the formation of immunity [32]. Lastly, as the presence of maternal autoimmune diseases are critical risk factor for the development of such diseases in offspring, we excluded the study population to children whose mothers had diagnoses with autoimmune diseases [33].

Baseline characteristics

For the pregnancy cohort, we identified 2,703,090 eligible children, among whom 1,516,574 (56.1%) were exposed to antibiotics in utero (Fig 1). Although restricted to children whose mothers were diagnosed with infection during pregnancy, proportion of cause of infection and proxies for the severity of infection (e.g., number of hospital visits with infection) were higher in the exposed group before PS weighting. Also, antibiotic exposure was more common among those with comorbidities of asthma and medication use (e.g., acetaminophen, NSAIDs) (Table 1). For the infancy cohort, among 3,347,049 eligible children with infections, 1,925,585 (57.5%) were exposed to antibiotics during early infancy. Before weighting, antibiotic-exposed infants were more likely to be male, have more causes of infection, and have mothers who were exposed to antibiotics during pregnancy (44.8% versus 39.4%) (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of cohort for pregnancy analysis before and after inverse probability of treatment weighting and sibling matching.

https://doi.org/10.1371/journal.pmed.1004677.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Baseline characteristics of cohort for early infancy analysis before and after inverse probability of treatment weighting and sibling matching.

https://doi.org/10.1371/journal.pmed.1004677.t002

After IPTW, we identified 1,336,754 exposed and 1,041,825 unexposed for pregnancy analysis, and 1,403,667 exposed infants and 1,277,387 unexposed infants for the infancy analysis, and all characteristics were well-balanced between the groups (aSMD < 0.1) (Table 1 and 2).

Exposure to antibiotics during pregnancy: IPTW and sibling-matched analyses

In pregnancy analysis, each group for respective outcome of interest was followed for a median duration of 7.6 years. The infection-restricted IPTW analyses showed null association between antibiotic exposure and autoimmune diseases in children, including type 1 diabetes (IPTW HR 1.14, 95% CI [0.96, 1.35]; p-value = 0.132), juvenile idiopathic arthritis (HR 1.02, 95% CI [0.85, 1.22]; p-value = 0.830), ulcerative colitis (HR 1.02, 95% CI [0.76, 1.37]; p-value = 0.895), Crohn’s disease (HR 1.16, 95% CI [0.98, 1.36]; p-value = 0.076), systemic lupus erythematosus (HR 0.70, 95% CI [0.49, 1.01]; p-value = 0.053), and Hashimoto’s thyroiditis (HR 1.06, 95% CI [0.91, 1.23]; p-value = 0.448).

Similarly, in the infection-restricted sibling-matched analyses, 263,401 exposed and 257,313 unexposed siblings were identified, and no associations were observed between antibiotic exposure during pregnancy and all 6 autoimmune diseases (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Risk of autoimmune disease associated with antibiotic exposure during pregnancy.

Abbreviations: HR, hazard ratio; PS, propensity score. *Incidence rate of respective outcome of interest was calculated per 100,000 person-years. ^†In the full cohort, crude analyses indicated a statistically significant increased risk of type 1 diabetes, ulcerative colitis, Crohn’s disease, and Hashimoto’s thyroiditis. However, when the analysis was restricted to individuals with infections and inverse probability of treatment weighting was applied, as well as in analyses restricted to exposure-discordant sibling pairs, all associations across the six autoimmune diseases were attenuated to null.

https://doi.org/10.1371/journal.pmed.1004677.g002

Exposure to antibiotics during pregnancy: Subgroup and sensitivity analyses

In the subgroup analysis by antibiotic spectrum within the pregnancy cohort, exposure to broad-spectrum antibiotics was associated with a modestly increased risk of Crohn’s disease (IPTW HR 1.19, 95% CI [1.00, 1.42]) (S3 Table). Elevated risks were also observed for Crohn’s disease in association with maternal exposure to cephalosporins (HR 1.29, 95% CI [1.07, 1.57]) and with exposure during the first and second trimesters (S4 and S7 Tables). No duration-dependent trends or effect modification by child sex were observed for any of the outcomes (S5 and S6 Tables). The results of the sensitivity analyses were consistent with the primary findings, except for Crohn’s disease, which showed an increased risk when exposure was redefined as two or more antibiotic prescriptions (S13 Table).

Exposure to antibiotics during early infancy: IPTW and sibling-matched analyses

In infancy analysis, each outcome of interest was followed for a median duration of 7.4 years. In the infection-restricted IPTW analyses, we found no difference in incidence of autoimmune disease risk between exposed and non-exposed groups for type 1 diabetes (IPTW HR 1.05, 95% CI [0.88, 1.26]; p-value = 0.594) and juvenile idiopathic arthritis (HR 1.11, 95% CI [0.93, 1.33]; p-value = 0.253), ulcerative colitis (HR 0.95, 95% CI [0.67, 1.36], p-value = 0.776), Crohn’s disease (HR 1.07, 95% CI [0.91, 1.25]; p-value = 0.403), systemic lupus erythematosus (HR 1.46, 95% CI [0.95, 2.26]; p-value = 0.087), and Hashimoto’s thyroiditis (HR 1.14, 95% CI [0.97, 1.33]; p-value = 0.104).

In the infection-restricted sibling-matched analyses, we included 421,725 exposed and 401,729 unexposed siblings for the infancy analysis. The sibling-matched HRs also consistently suggest no association between antibiotic exposure during early infancy and all 6 autoimmune diseases (Fig 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Risk of autoimmune disease associated with antibiotic exposure during early infancy.

Abbreviations: HR, hazard ratio; CI, confidence interval; PS, propensity score. *Incidence rate of respective outcome of interest was calculated per 100,000 person-years. ^†In the full cohort, crude analyses indicated a statistically significant increased risk of juvenile idiopathic arthritis, ulcerative colitis, Crohn’s disease, and systemic lupus erythematosus. However, when the analysis was restricted to individuals with infections and inverse probability of treatment weighting was applied, as well as in analyses restricted to exposure-discordant sibling pairs, all associations across the six autoimmune diseases were attenuated to null.

https://doi.org/10.1371/journal.pmed.1004677.g003

Exposure to antibiotics early infancy: Subgroup and sensitivity analyses

In the IPTW analyses, no associations were observed by antibiotic spectrum, individual subclasses, or duration-dependent trends for any of the outcomes (S8, S9 and S10 Tables). However, the risk of Hashimoto’s thyroiditis was increased among males (IPTW HR 1.40, 95% CI [1.02, 1.90]) and following antibiotic exposure within two months after birth (HR 1.30, 95% CI [1.07, 1.58]) (S11 and S12 Tables). The findings from sensitivity analyses demonstrated substantial consistency with our primary results (S14 Table).